DELE CA2 Part A - GAN

Background Research

CIFAR-10 is a dataset which comprises 60,000 32x32 colour images in 10 classes/categories, with 6000 images per class. There are 50000 training images and 10000 test images in total (i.e. 5000 training and 1000 test images per class).

The categories are:

  1. airplane
  2. automobile
  3. bird
  4. cat
  5. deer
  6. dog
  7. frog
  8. horse
  9. ship
  10. truck

The classes are completely mutually exclusive. There is no overlap between automobiles and trucks. "Automobile" includes sedans, SUVs, things of that sort. "Truck" includes only big trucks. Neither includes pickup trucks.

This ensures that each class is isolated from the rest.

Since we have been tasked with image generation, GANs come to mind. There are a couple of network architectures we could employ:

Due to time constraints, I was only able to try out DCGAN.

Import General Dependencies

Declare Global Constants

General Utilities

EDA

DCGAN

GAN Components

Visualise Model Architecture

Airplanes

Automobiles

Birds

Cats

Deer

Dogs

Frogs

Horses

Ships

Trucks

Evaluation

Conclusions

References